Thèse soutenue

Méthodes d'apprentissage à noyau pour l'estimation distribuée dans les réseaux de capteurs sans fil

FR  |  
EN
Auteur / Autrice : Mehdi Essoloh
Direction : Cédric RichardHichem Snoussi
Type : Thèse de doctorat
Discipline(s) : Optimisation et sûreté des systèmes
Date : Soutenance en 2008
Etablissement(s) : Troyes
Ecole(s) doctorale(s) : Ecole doctorale Sciences pour l'Ingénieur (Troyes, Aube)

Résumé

FR  |  
EN

This thesis proposes a new frame for estimation problems in wireless sensor networks thanks to learning methods built on reproducing kernels. In a first part, our work deals with the sensor network localization problem thanks to reproducing kernel Hilbert space formalism. While respecting energy constraints and limitations in computation capabilities, coordinate estimation is executed thanks to range measurements between sensors and a priori known locations of some small fraction of deployed sensors. By considering these dissimilarities as elements of a Gram matrix, we investigate two distributed approaches: one is related to the pre-image problem, widely used in denoising applications, the other one is based on a kernel matrix regression approach, recently introduced in bio-engineering. In a second part, we propose a distributed learning strategy for temperature field estimation in wireless sensor networks. We note that sparse approximation, enabling an efficient control of the order model, holds with algorithmic constraints of wireless sensor networks. Our work is based on non-linear adaptive filtering techniques with kernels and we demonstrate its relevant use for distributed regression problem in wireless sensor networks